Latent anxiety and depression dimensions differ amongst patients with eating disorders: A Swedish nationwide investigation

Abstract Objective Anxiety and depression symptoms are common in individuals with eating disorders. To study these co‐occurrences, we need high‐quality self‐report questionnaires. The 19‐item self‐rated Comprehensive Psychopathological Rating Scale for Affective Syndromes (CPRS‐S‐A) is not validated in patients with eating disorders. We tested its factor structure, invariance, and differences in its latent dimensions. Method Patients were registered by 45 treatment units in the Swedish nationwide Stepwise quality assurance database for specialised eating disorder care (n = 9509). Patients self‐reported their anxiety and depression symptoms on the CPRS‐S‐A. Analyses included exploratory and confirmatory factor analyses (CFA) in split samples, and testing of invariance and differences in subscales across eating disorder types. Results Results suggested a four‐factor solution: Depression, Somatic and fear symptoms, Disinterest, and Worry. Multigroup CFA indicated an invariant factor structure. We detected the following differences: Patients with anorexia nervosa binge‐eating/purging subtype scored the highest and patients with unspecified feeding and eating disorders the lowest on all subscales. Patients with anorexia nervosa or purging disorder show more somatic and fear symptoms than individuals with either bulimia nervosa or binge‐eating disorder. Conclusion Our four‐factor solution of the CPRS‐S‐A is suitable for patients with eating disorders and may help to identify differences in anxiety and depression dimensions amongst patients with eating disorders.


| Measurement issues
To study co-occurring symptoms among eating disorders, anxiety, and depression, we need high-quality measures of symptoms to delineate differences in eating disorder presentation in clinical and population samples. To assesses anxiety and depression symptoms, measures like the Patient Health Questionnaire-9 (PHQ-9; Kroenke et al., 2001) the Generalised Anxiety Disorder Assessment (GAD-7; Spitzer et al., 2006), the Symptom Checklist-90 (SCL-90;Fittig et al., 2008) and its short form, the Brief Symptom Inventory (BSI; Beesdo-Baum et al., 2012) are widely used in research. The PHQ-9 and GAD-7 factor structures and scores have been validated in samples of patients with eating disorders and the general population, indicating that both questionnaires are suitable (Wisting et al., 2021).
However, as PHQ-9 and GAD-7 are strictly based on diagnostic criteria, these questionnaires only cover a limited range of anxiety and depression symptoms. The SCL-90 and BSI cover a wider range of symptoms but are less frequently used in eating disorder research.
Additionally, studies often only calculate global summary scores, ignoring more fine-grained information of subscales. Hence, in order to explore the whole spectrum of symptoms, broader assessment tools to better understand the heterogeneity in eating disorder presentations, evading the cost-and time-related limitations of diagnostic interviews, are urgently needed. One potential scale of interest is the Comprehensive Psychopathological Rating Scale (CPRS; Asberg & Schalling, 1979) which consists of 65 items and was originally developed to evaluate treatment outcomes in psychological interventions. The scale includes items covering symptoms of psychiatric disorders, such as schizophrenia but also anxiety and depression. The scale was originally developed in Sweden, and has been translated into most other European languages. The complete version of the CPRS is rarely used, but shorter subscales have been deemed to be more useful, such as the Montgomery Åsberg Depression Rating Scale (MADRS; Montgomery & Asberg, 1979) and the Self-rating Scale for Affective Syndromes (CPRS-S-A; Svanborg & Asberg, 1994). The latter is in focus here, and is designed to contain subscales for depression, anxiety, and compulsivity.
A previous analysis of the CPRS-S-A questionnaire in a subsample of the data available for our investigation compared a global CPRS-S-A score across patients with different eating disorders. Results showed that patients with an unspecified feeding or eating disorder reported fewer problems than patients with other eating disorders. Additionally, patients with the anorexia nervosa bingeeating/purging subtype reported more problems compared with atypical anorexia nervosa patients (Ekeroth et al., 2013). One issue of the questionnaire is the construction of its three subscales. When calculating the subscales, it is advised to include the same item in several subscales. Therefore, the subscales are highly correlated. In the previous analyses the correlations ranged from 0.78 to 0.86 and are inflated, rendering the original subscales unreliable. Therefore, in this study, we investigated differences in depression and anxiety dimension amongst patients with eating disorders using newly derived subscales of the Self-rating Scale for Affective Syndromes (CPRS-S-A); a short form of the CPRS (Svanborg & Asberg, 1994) in one of the world's largest clinical sample of more than 9000 patients with eating disorders in Sweden.

| Sample
The sample comprises inpatients and outpatients registered by 45 treatment units in the Stepwise quality assurance database for specialised eating disorder care in Sweden aged 18 years and older  Stepwise is a nationwide internet-based data collection system, which includes individuals through medical or selfreferral, if intention to treat has been established, and if the individual received a formal eating disorder diagnosis (Birgegård et al., 2022). The database has been used since 2005 and our data were extracted on November 23, 2017. At data extraction, approximately 10,470 adult entries had been registered (Supplementary   Table S1). Depending on the patient's endorsement of binge eating or purging in either the Eating Disorder Examination questionnaire (Luce & Crowther, 1999) or the Structured Eating Disorder Interview (de Man Lapidoth & Birgegård, 2010), we re-assigned DSM-5 diagnoses.

| Eating disorder diagnosis
We used 18.5 kg/m 2 as the cutoff value for underweight in anorexia nervosa. Anorexia nervosa without weight criterion (n = 50) or without amenorrhea (n = 186) that had a body mass index (BMI) lower than 18.5 kg/m 2 who endorsed any binge eating or purging were assigned anorexia nervosa binge-eating/purging. If none endorsed, they were assigned an anorexia nervosa restricting diagnosis (n without weight criterion = 84 or n without amenorrhea = 144). If their BMI was above 18.5 kg/m 2 , we assigned an atypical anorexia nervosa diagnosis (n without weight criterion = 441 or n without amenorrhea = 402). Eating Disorder Not Otherwise Specified type 3 or bulimia nervosa without sufficient duration/frequency criteria (n = 833) was assigned as bulimia nervosa diagnosis because those criteria are relaxed in DSM-5. Further, Eating Disorder Not Otherwise Specified example 4 was kept as 'purging disorder'. The remaining unspecified eating disorders that were not classified into either of these categories were termed 'unspecified feeding or eating disorder' (UFED), consisting of patients with 'chewing and spitting', bulimia nervosa/ binge-eating disorder with low frequency/duration, or other residual types that did not fit any of the main categories (Supplementary Table S1).

| Exclusion
We excluded 801 duplicated entries of repeated registrations of the same individual, keeping the first registration. Subsequently, we iteratively excluded two individuals with missing age, 16 not assigned a treatment centre, 120 without a clinical eating disorder diagnosis, and 22 because they had not answered the CPRS questionnaire. The final sample comprised 9509 patients with eating disorders.

| Ethics
When patients were entered into the database, clinicians recorded consent for general research use of their data and 3% declined participation. This study is approved by the Stockholm Regional Ethics Board (Reg. no. 2009/196-31/4).

| Comprehensive Psychopathological Rating Scale, self-rated version for Affective Syndromes
At registration, the patients answered 19 items of the CPRS-S-A. We present the instrument as Supplementary Material. The answer options are different for each question, but they are on a scale from 0 to 3, rated in 0.5-point increments. We recoded these values to 0-6. We renamed item 19, titled 'Zest for life' in the MADRS-S to 'Suicidal thoughts' to represent its content better.

| Exploratory factor analyses
We calculated pairwise Pearson correlations amongst all items ( Figure 1) in the full sample (n = 9509). We inspected the matrix visually for singularity, multicollinearity, and redundancy of items (i.e., values <0.30 and >0.90). We calculated the determinant of the matrix (Dziuban & Shirkey, 1974), the Kaiser-Meyer-Olkin (KMO) statistic (Kaiser, 1974), and performed Bartlett's Test of Sphericity (Bartlett, 1950), to test if our data are suitable for an exploratory HÜBEL ET AL.
-3 of 13 factor analysis. To inform our decision on the underlying factor structure, we performed parallel analysis (Horn, 1965), and calculated the Very Simple Structure criterion (VSS; Revelle & Rocklin, 1979), and Velicer's Minimum Average Partial criterion (Velicer, 1976). We performed the exploratory factor analysis on 70% (n = 6656) of the sample using the maximum likelihood estimator in the 'psych' R package (Revelle & Revelle, 2015). Given that the CPRS items have seven answer options, we treated them as continuous. We allowed the factors to correlate using oblimin rotation. To judge the fit of our model, we applied the criteria as outlined in Table 1 (Hu & Bentler, 1999). We retained the items with factor loadings of >0.30. If multiple models showed adequate fit, we would choose the model with factors that encompass the greatest number of items.

| Confirmatory factor analysis and factor scores
We validated our exploratory factor analysis model with a confirmatory factor analysis (CFA) on the remaining 30% participants using the 'lavaan' R package (Rosseel, 2012). We interpreted fit statistics F I G U R E 1 Pairwise Pearson's correlations amongst the Self-rating Scale for Affective Syndromes (CPRS-S-A) items. We calculated the correlations in 9509 participants registered in Stepwise, the Swedish clinical eating disorder database. We estimated the number of independent traits in the matrix using the Galwey method and adjusted the α threshold (α = 0.003) accordingly. All correlations are statistically significant at this α threshold. Saturation represents the strength of the correlation. Positive correlations are red.
T A B L E 1 Criteria for a good fit (Hu & Bentler, 1999) Root mean square error of approximation (RMSEA)

≤0.05
Bayesian information criteria (BIC) Smaller than other models (Hu & Bentler, 1999;Schreiber et al., 2006) and considered a Comparative Fit Index (CFI) ≥0.95 as good fit. Subsequently, we computed the CFA in the full sample (n = 9509) to provide fit statistics and calculate factor scores, using the Bartlett estimator for continuous items.

| Descriptive indices and psychometric properties
We show responses to the individual items as frequency plots ( Figure 2) and distributions of the factor scores as histograms and qq plots for the complete sample (Supplementary Figure S1) while presenting box plots per eating disorder (Figure 4). We also report mean and standard deviations for our generated factor scores and report Cronbach's α (Bland & Altman, 1997;Cronbach, 1951) and McDonald's ω (Hayes & Coutts, 2020) as measures of internal consistency.

| Multigroup confirmatory factor analysis
We performed a multigroup confirmatory factor analysis (MGCFA) to test if the questionnaire elicits the same responses, response patterns, and has the same underlying factor structure across eating disorder diagnostic groups. If statistical invariance in responding is found, then we can compare scores and subscale scores across groups. Different types of measurement invariance exist: configural, the factor structure is similar across groups; metric, factor loadings are similar across groups; scalar, intercepts (i.e., group means) are similar; and strict, residuals (i.e., variances) are similar across the groups. We tested for these invariance models in a stepwise procedure from the least restricted model to the fully restricted model.
Overall, invariance indicates that different groups are from the same population.

| Group comparisons
We judged the distribution of the factor scores by visually inspecting qq and distribution plots (Supplementary Figure S1). None of the four subscales showed a normal distribution. Therefore, we performed non-parametric Kruskal-Wallis one-way ANOVAs. If significant, Dunn's post-hoc tests were carried out with a Benjamini Hochbergadjusted level of significance for the pairwise comparisons. F I G U R E 2 Endorsement of the Comprehensive Psychopathological Rating Scale Self-rating Scale for Affective Syndromes (CPRS-S-A) 19 items version in the Stepwise sample (n = 9509). The saturation of blue indicates a higher endorsement on the specific item. We display percentages. The answer options differed across items, with higher values indicating a stronger endorsement.

| Convergent and divergent validity
HÜBEL ET AL.

| Descriptives of the CPRS-S-A in stepwise
The CPRS-S-A showed a Cronbach's α (α = 0.90) and McDonald's ω (ω = 0.92) in our sample (Supplementary Table S2). The distributions of answers to the questionnaire items are displayed in Figure 2 for the full sample and Supplementary Table S3 for the discovery sample.

| Suitability of the data for factor analysis
Prior to factor analyses, the suitability of the data was investigated.
None of the items showed zero or near-zero variance (Supplementary analyses. Pearson's correlations ranged from 0.09 to 0.68 (Figure 1).
The exploratory factor analysis was conducted on one random split of the sample (n = 6656; 70%). As we were primarily interested in core anxiety and depression symptoms, we excluded the items '14.
Furthermore, they loaded strongly on one factor by themselves, representing an index of compulsion. If these items had remained in the model, they would have lowered our power to measure meaningful underlying factors as they would have distorted the model towards their own factor. We, furthermore, excluded the item '11. Health concerns', because its correlation with the other items was small (r = 0.09-0.29; Figure 1), rendering it unsuitable for factor analysis. Cronbach's α remained stable after these items were dropped (Supplementary Table S6).

| Exploratory factor analysis
Very simple structure (Supplementary Table S7 Tables S9-13). As factors were considered to be correlated, factors were realigned using an oblique rotation. The factor loadings for each item, after rotation, are listed in Figure 3. Items 3 and 4 (Irritation and anger, and Sleep, respectively) did not load on any of the factors and are therefore not included in the confirmatory factor analysis.
We labelled the four factors: F1 Depression, F2 Somatic and fear symptoms, F3 Disinterest, and F4 Worry.

| Confirmatory factor analysis
We conducted the CFA in the remaining 30% of the sample (n = 2853; Supplementary Note: The cut off for each statistic to signify good fit is listed in each header (Hu & Bentler, 1999). The model with the lowest BIC is preferred, Cumulative variance indicates the part of the total variance explained by all items comprising the factors. The factor analysis was performed on 16 items of the Comprehensive Psychopathological Rating Scale Self-rating Scale for Affective Syndromes (CPRS-S-A) in the Swedish quality register for eating disorder care, Stepwise (n = 6656). Model in bold was chosen as the best fitting model.

| Multigroup confirmatory factor analysis
Our multigroup confirmatory factor analysis resulted in full configural and metric invariance, indicating that the factor structure and the factor loadings are comparable across eating disorders (Supplementary Table S15). Furthermore, the questionnaire showed partial scalar invariance when freeing up the intercepts of item five and eight, meaning that the means were similar across groups apart from item five (less appetite) and eight (less motivation).

| Factor scores
We calculated factor scores for each individual based on the final model ( Figure 4 and Supplementary Table S16). We compared the F I G U R E 3 Exploratory factor analysis of 16 items of the Comprehensive Psychopathological Rating Scale Self-rating Scale for Affective Syndromes (CPRS-S-A). The path diagram shows item factor loadings and between-factor correlations for the four factors of Depression, Somatic and fear symptoms, Disinterest, and Worry. Paths with a factor loading of <0.3 were omitted.
-7 of 13 factor scores using Kruskal-Wallis one-way analysis of variance, and Dunn's post hoc test (Table 3)

| Context of existing literature
Difference to original scale. Our factor structure differs substantially from the original CPRS-S-A (Svanborg & Asberg, 1994  the original CPRS-S-A that combined worry symptoms with the somatic and fear-based symptoms. Overall, our analyses suggest a substantially different factor structure compared with the original. General differences amongst patients with eating disorders. We explored differences in the new subscales amongst patients with eating disorders. On the one hand, comparisons suggested that patients with anorexia nervosa binge-eating/purging score higher on all four subscales, consistent with the previous report based on a subsample of our analysis (Ulfvebrand et al., 2015). On the other hand, patients with unspecified feeding and eating disorders had the lowest scores across all four subscales in line with their subsyndromal expression of eating disorders.
Specific differences. In addition to these overarching differences, we detected differences for specific factors. On factor 2 Somatic and fear symptoms, patients with anorexia nervosa or purging disorder scored higher than individuals with either bulimia nervosa or bingeeating disorder. These differences may indicate that the somatic complications seen in anorexia nervosa (Westmoreland et al., 2016) and purging disorder may be captured by items on this factor summarising somatic fear symptoms. Furthermore, patients with anorexia nervosa and purging disorders may perceive these somatic and fear symptoms more strongly than patients with bulimia nervosa or bingeeating disorder. Fear has been proposed as a fundamental mechanism in the development of anorexia nervosa (Murray et al., 2018).
Depression and anxiety are risk factors for eating disorders (Meier et al., 2015;Steinhausen et al., 2015), but certain symptoms of anxiety or depression may represent somatic or psychiatric complications or sequelae of the eating disorder itself. However, in some cases, depressive and anxiety symptoms may be independent of the eating disorder. This underscores the importance of investigating anxiety and depression on the dimension or symptom level rather than using total scores.

| Limitations
Our study may be biased due to limitations. The sample consisted predominantly of women which limits the ability to identify sex differences. Eating disorders are more commonly diagnosed amongst women, however, men are underrepresented in eating disorder research. This may be due to a lack of awareness and understanding for these disorders among the wider community and clinicians or may represent an underlying true sex difference. Our sample included Swedish treatment seeking patients of mostly white European ancestry limiting the generalisability of our findings. Furthermore, patients in healthcare registers may represent a more severe subpopulation of individuals with eating disorders.
Hence, the factor structure and our observed differences amongst F I G U R E 5 Correlation matrix of CPRS-S-A dimensions with external correlates. The correlation matrix shows Pearson's correlations of the original CPRS-S-A anxiety and depression dimensions, the newly derived 4 factor solution, the Clinical Impairment Assessment (CIA) total score, and both subscales of the Structural Analysis of Social Behaviour (SASB) self-affirmation and self-control als well as height. Sample sizes range between 8146 and 9509. Positive correlations are red and negative correlations are blue.
patients with eating disorders may not replicate across other ancestry or cultural groups or in individuals with less severe presentations. Our analyses were cross-sectional and did not include a comparison group without eating disorders or any psychiatric disorder.

| Future directions
To address a few of our limitations, future studies should confirm our newly detected factor structure in community samples, samples with other psychiatric disorders, and include a healthy comparison group.
Optimally, researchers would collect repeated measures of the CPRS-S-A that would further our understanding of how these constructs develop over time and how levels of depression, disinterest, fear, and worry may change with treatment. Future studies could investigate clinical cut-offs to measure comorbid depressive and anxiety disorders.

| Conclusions
In summary, our four-factor solution of the CPRS-S-A is suitable for adult patients with different eating disorders and identifies differences in anxiety and depression dimensions. An easily administered,